Today I was benchmarking my upcoming PhpFaces code and was thoroughly impressed by the numbers returned by apachebench; until I started testing concurrency, that is. I tested a simple phpinfo() call with various levels of concurrency and was shocked to see that the problem may not be my code, but some bottleneck within either apache, php, or hardware. To test easy things (http server) first, I decided to benchmark using lighttpd, which I use on my local workstation. My perception of lighttpd+fastcgi has always been that since it's CGI it is slow--that's what they taught us, right? Well, FastCGI works differently than normal CGI, but wikipedia says it better than I can :).

To cut straight to the numbers:

Server	What?	# of Requests	Concurrency	Min Time	Max Time
Apache	phpinfo()	500	500	776	904
Lighttpd	-	-	-	377	429
Apache	PhpFaces	500	500	12517	40226
Lighttpd	-	-	-	1844	9957
Apache	PhpFaces (realistic)	150	25	1802	4060
Lighttpd	-	-	-	1595	2824

Complete apachebench output + server specs can be found here (self hosted, may not always be available).

I must say that I was taken completely by surprise by these results. While Lighttpd may not be the best for all sites, for the vast majority of sites that are either self-hosted or otherwise on dedicated installation of apache it is a shining winner. Lighttpd has all the basic functionality of apache including mod_rewrite, mod_alias, mod_redirect, mod_auth and mod_cgi/fastcgi. There are many others including mod_proxy, but I have yet to experiment with that. More information about Lighttpd can be found at their website, here.